State-of-the-artsequencelabelingsystemstraditionallyrequirelargemountsoftask-specificknowledgeintheformofhand-craftedfeaturesanddatapre-processing.Inthispaper,weintroduceanovelneu-tralnetworkarchitecturethatbenefitsfrombothword-andcharacter-levelrepresenta-tionsautomatically,byusingcombinatio